Industry Leaders Call for Slowing the AI Race

Dario Amodei, CEO of Anthropic and one of the AI industry’s leading figures, has called for slowing the pace of improvements to model capabilities, giving safety and oversight mechanisms time to catch up with technological development. In an approximately 3,800-word essay published yesterday (Shabbat), he warned that development is moving faster than researchers’ ability to understand the systems and ensure they remain under control.

Amodei wrote: "In recent months, I have become convinced that fully addressing the risks requires even greater caution—not only investing in preventing them, but also slowing the pace of capability improvements so that prevention work has time to catch up." He later clarified: "We must slow the pace at which we improve the capabilities of AI models. Progress will still look fast, and we must use the time we gain wisely."

Amodei is not proposing that research be stopped or model training be frozen. Instead, he is calling for a slowdown of one or two years, during which oversight mechanisms could be improved, what is happening inside the systems could be better understood, and tests that models cannot deceive could be developed. He said that while such a slowdown seemed unnecessary to him in 2023, today "the picture is completely different."

At the heart of the concern is the fact that AI systems are already helping develop the next generations of AI systems, a process that could accelerate progress beyond humanity’s ability to understand and control it. Amodei cited experiments and incidents in which AI agents went beyond testing environments and acted in ways they had not been asked to act. According to him, even early versions of Anthropic’s "Claude" broke into external systems, and the incidents were discovered only afterward.

In response to the risks, he proposes permanent outside auditors at companies, mandatory safety standards, and cooperation among democratic countries, while preserving America’s technological advantage over China. From his perspective, the central threat is the gap between the speed of development and the ability to oversee it.

Amodei is not the only senior figure in the AI industry who has recently raised the possibility of hitting the brakes, after years in which the race focused mainly on accelerating development. OpenAI CEO Sam Altman said as early as July that it may be time to slow down, to give human society time to prepare for the changes the technology brings. In recent days, he also told company employees that OpenAI is prepared to consider reducing the pace of progress, in coordination with other AI labs.

Bill Gates has also joined the warnings. Last month, he cautioned that AI would be "the greatest equalizer ever invented, or the worst source of injustice," arguing that the world is still unprepared for the risks accompanying the technology, including job losses, cyberattacks, biological weapons, and even the possibility of losing control. Gates wrote that he would support a credible plan for a global slowdown, but estimated that economic and geopolitical pressures would make such a move very difficult to put together.

Reactions to Amodei’s essay came quickly. Elon Musk offered a short, sharp message: "Dario is right." Shortly afterward, Altman joined the call as well. "I agree with Dario that we need to slow progress at the frontier," he tweeted, adding that the issue has been at the center of discussions at OpenAI in recent weeks.